How We Learn: Why Brains Learn Better Than Any Machine . . . for Now

general powers of reasoning and abstraction. Artificial neural networks neglect an essential point: human learning is not just the setting of a pattern-recognition filter, but the forming of an abstract model of the world.

LEARNING IS OPTIMIZING A REWARD FUNCTION

The immense number of parameters that neural networks possess often leads to a second obstacle, which is called “overfitting” or “overlearning”: the system has so many degrees of freedom that it finds it easier to memorize all the details of each example than it is to identify a more general rule that can explain them.

At any rate, working memory never lasts more than a few seconds: as soon as we get distracted by something else, the assembly of active neurons fades away.

Go. The problem is colossal for a simple reason: it is only at the very end that the system receives a single reward signal, indicating whether the game was won or lost. During the game itself, the system receives no feedback whatsoever—only the final checkmate counts.

LEARNING IS ADJUSTING THE PARAMETERS OF A MENTAL MODEL

As a neuron learns, its very shape changes too. A mushroom-like structure called a “dendritic spine” forms at the place on the dendrite where the synapse lands. If necessary, a second synapse emerges to double the first. Other synapses that land on the same neuron are also strengthened.18 Thus, when learning is prolonged, the very anatomy of the br

Nico, a young painter who, while visiting the Marmottan Museum in Paris, managed to make an excellent copy of Monet’s famous painting Impression, Sunrise

The ideal scenario is to offer the guidance of a structured pedagogy while encouraging children’s creativity by letting them know that there are still a thousand things to discover.